Reviewed PR #348 (fk/feat/histo-peso-leopard) against merge base 0f40a2e501624edaae49b2736e5fe70fadb85ace without editing the monorepo.
Read the thermo-nuclear review standard, the vxData ingestion skill, the histo knowledge index, the full branch diff, and the canonical vxData schema and SDK implementations.
Verified the public LEOPARD S3 inventory shape and multipart ETag assumptions against the live public bucket listing.
Ran targeted Ruff and ty checks for f_20260824_peso_leopard; both passed. The app-wide typecheck remains red in unrelated legacy jobs.
Requested changes for mutation-before-conflict-validation, missing post-index/update-free patient verification, conflicting S3 alias defaults, and job-local duplication of canonical SDK batching.
Lessons Learned: Pitfalls
The app-wide typecheck emits a very large legacy error set and is not useful for isolating this job. Targeted ty check should be run first for review work in this app.
The older vxData processing skill still contains some historical API guidance, so current SDK source must remain the canonical reference for concrete client behavior.
Lessons Learned: Improvements
The vxData ingestion SOP would benefit from an explicit discover -> read-only preflight -> apply -> post-write exact verification pattern, including how to prove an update-free reconciliation pass.
Document that SDK namespace creates perform read-after-write and that bulk read chunking must stay in the SDK. This would prevent ingestion jobs from rebuilding transport batching locally.
Centralize the documented default VXDATA_S3_ALIAS for transfer and indexing jobs to prevent operator-specific alias names from entering code.
